Multi-Disease Detection and Classification of Sugarcane Leaves using Transfer Learning

The early detection of multiple diseases in sugarcane is crucial for improving crop yield and reducing losses. Traditional manual inspection methods are time-consuming, errorprone, and require expert knowledge. In this study, we propose a novel approach for detecting multiple diseases in sugarcane leaves using a combination of transfer learning and a custom hybrid model. Initially, pre-trained models like MobileNet and DenseNet are fine-tuned on a sugarcane disease dataset to leverage the feature extraction capabilities of transfer learning. We then propose a new model, the MobileNet-DenseNet Concatenation (MobileNet-DenseNet Concat), which combines the strengths of both architectures. This hybrid model fuses the lightweight, efficient nature of MobileNet with the deep feature extraction capabilities of DenseNet, creating a robust model for multidisease classification. The model is trained and evaluated on a labeled sugarcane disease dataset, and its performance is compared with individual transfer learning models. The proposed model achieves improved $96 \%$ accuracy, precision, recall, and F<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">1</inf> scores, outperforming existing approaches in detecting multiple diseases simultaneously. Additionally, techniques like bounding boxes are used to localize disease regions on the leaves. The results demonstrate that the MobileNet-DenseNet Concat model offers a promising solution for automating sugarcane disease detection, potentially benefiting farmers and agronomists by facilitating early intervention and treatment strategies.